Fast 4D-STEM-Based Phase Mapping for Amorphous and Mixed Materials

Area of Science:

• Materials Science
• Data Analysis
• Electron Microscopy

Background:

• 4D-scanning transmission electron microscopy (4D-STEM) excels with crystalline materials.
• Analyzing amorphous and mixed materials in 4D-STEM is challenging due to phase separation difficulties.
• Nonnegative matrix factorization (NMF) offers potential for phase separation but is hindered by large datasets and computational complexity.

Purpose of the Study:

• To accelerate the application of NMF for phase mapping in large 4D-STEM datasets, particularly for amorphous and mixed materials.
• To overcome the computational challenges associated with NMF's iterative algorithms and stopping conditions.
• To enable structure-independent phase separation and mapping in complex materials.

Main Methods:

• Utilized QB decomposition for randomized NMF (RNMF) to drastically accelerate data factorization.
• Applied RNMF to large 4D-STEM datasets, significantly reducing iteration time.
• Employed principal component analysis (PCA) for initial exploratory data analysis and dimensionality assessment.

Main Results:

• Demonstrated drastically accelerated factorization of large 4D-STEM datasets using RNMF.
• Successfully enabled structure-independent phase mapping on extensive 4D-STEM data.
• Validated the RNMF approach on synthetic (ZrCuAl) and real-world samples (TiO2/SiO2, Li-ion cathode interface).

Conclusions:

• RNMF provides a computationally efficient method for phase mapping in 4D-STEM, applicable to amorphous and mixed materials.
• The accelerated approach overcomes limitations of traditional NMF, allowing analysis of larger and more complex datasets.
• This technique enhances the interpretability of 4D-STEM data for diverse material systems.